Wyniki wyszukiwania - BazTech

1

Stochastic schemata exploiter-based optimization of hyper-parameters for XGBoost

Makino Hiroya, Kita Eisuke

Computer Assisted Methods in Engineering and Science

|

2024

|

Vol. 31, no. 1

113--132

EN

XGBoost is well-known as an open-source software library that provides a regularizing gradient boosting framework. Although it is widely used in the machine learning field, its performance depends on the determination of hyper-parameters. This study focuses on the optimization algorithm for hyper-parameters of XGBoost by using Stochastic Schemata Exploiter (SSE). SSE, which is one of Evolutionary Algorithms, is successfully applied to combinatorial optimization problems. SSE is applied for optimizing hyper-parameters of XGBoost in this study. The original SSE algorithm is modified for hyper-parameter optimization. When comparing SSE with a simple Genetic Algorithm, there are two interesting features: quick convergence and a small number of control parameters. The proposed algorithm is compared with other hyper-parameter optimization algorithms such as Gradient Boosted Regression Trees (GBRT), Tree-structured Parzen Estimator (TPE), Covariance Matrix Adaptation Evolution Strategy (CMA-ES), and Random Search in order to confirm its validity. The numerical results show that SSE has a good convergence property, even with fewer control parameters than other methods.

2

Application of grammatical evolution to stock price prediction

Kita E., Sugiura H., Zuo Y., Mizuno T.

Computer Assisted Methods in Engineering and Science

|

2017

|

Vol. 24, no. 1

67--81

EN

Grammatical evolution (GE) is one of evolutionary computation techniques. The aim of GE is to find the function or the executable program or program fragment that will find the optimal solution for the design objective such as the function for representing the set of given data, the robot control algorithm and so on. Candidate solutions are described in bit string. The mapping process from the genotype (bitstring) to the phenotype (function or program or program fragment) is defined according to the list of production rules of terminal and non-terminal symbols. Candidate solutions are evolved according to the search algorithm based on genetic algorithm (GA). There are three main issues in GE: genotype definition, production rules, and search algorithm. Grammatical evolution with multiple chromosomes (GEMC) is one of the improved algorithms of GE. In GEMC, the convergence property of GE is improved by modifying the genotype definition. The aim of this study is to improve convergence property by changing the search algorithm based on GA with the search algorithm based on stochastic schemata exploiter (SSE) in GE and GEMC. SSE is designed to find the optimal solution of the function, which is the same as GA. The convergence speed of SSE is much higher than that of GA. Moreover, the selection and crossover operators are not necessary for SSE. When GA is replaced with SSE, the improved algorithms of GE and GEM Care named “grammatical evolution by using stochastic schemata exploiter (GE-SSE)” and “grammatical evolution with multiple chromosome by using stochastic schemata exploiter (GEMC-SSE)”, respectively. In this study, GE-SSE is compared with GE in the symbolic regression problem of polynomial function. The results show that the convergence speed of GE-SSE is higher than that of original GE. Next, GE-SSE and GEMC-SSE are compared in stock price prediction problem. The results show that the convergence speed of GEMC-SSE is slightly higher than that of GE-SSE.

3

Improvement of evolutionary algorithm based on schema exploiter

Maruyama T., Kita E.

Computer Assisted Mechanics and Engineering Sciences

|

2008

|

Vol. 15, No. 2

85-98

EN

Stochastic Schemata Exploiter (SSE) is one of the evolutionary optimization algorithms for solving the combinatorial optimization problems. We present the Extended SSE (ESSE) algorithm which is composed of the original SSE and new ESSE operations. The ESSE is compared with the original SSE, simple genetic algorithm (SGA), and GA with Minimal Generation Gap (MGG) in some test problems in order to discuss its features.